Deterministic first. AI where it earns its place.
Context
Security operations fail at the seams — untested rules pushed to production, playbooks nobody can explain, "intelligent" automation that can't be audited when it matters. The cost of a wrong automated action in a SOC is not an inconvenience; it's an incident.
Decision
Orchestration is deterministic by default. Non-deterministic steps — LLM enrichment, agentic triage — are isolated behind explicit boundaries, given bounded authority, and made reviewable by a human before consequential action. Every significant choice is written down as a decision record your team owns after we leave.
Consequences
Automation you can test before deployment, trace after it, and defend in an audit. Slower to promise, faster to trust.